The prediction of science success at High School Entrance Exam with artificial neural network

In this study, it is aimed to investigate the level of predicting the success of Science in high school entrance exam (LGS) by using artificial neural networks (YSA) by associating the success of students in Science classes from 4th grade elementary school to the higher-class level. For this purpose, Pearson Moment Product Correlation analysis results were analysed in SPSS program in order to examine the relationship between the academic achievement of Science Grade and LGS Science Sub-Test achievement of 4th, 5th, 6th, 7th and 8th grade students. In MATLAB program, artificial neural network modelling performance was examined to understand the level of predicting. Data sets used in the research were obtained from the Ministry of Education E-school system for 1027 students who graduated from 24 schools in 17 districts of Bursa in the 2017-2018 academic year and also entered 2018 LGS, and their marks from the 4th to 8th grade and LGS result documents that do not contain personal information. In the scope of research, when the correlations between the LGS Science Sub-Test and Science test success were examined (p <0.001), the highest and lowest correlations were found the 8th grade science exams (r = 0.70) and the 4th grade science exams (r = 0.57), respectively. According to this result, it can be said that the medium and high-level relationship between the classes is related with the fact that the science course curriculum is spiral and holistic structure from 4th grade to 8th grade. The highest performance values were found as Learning R = 0.8059, Verification R = 0.7408, Test R = 0.7568, RMSE = 2.3504 at the network architecture which was generated in the second sub-problem and the trained network with 845 student data. Using the data of 182 students, it was obtained a high-level relationship with r=0.7506 (p<0.001), while the correct numbers of simulated real LGS Science Sub-Test compared with the correct numbers predicted by the computer after the training process of the network. In addition, considering the predictive performance of the network in 20 questions, it was obtained that 124 students in 182 students were correctly estimated in [+ 2, -2] error value range.